Flood Frequency Analysis Using the

Gumbel Distribution


Never Mujere
University of Zimbabwe,
Department of Geography and Environmental Science,
Box MP 167, Mount Pleasant, Harare
E-mail mujere@arts.uz.ac.zw

Abstract
This paper presents results of a study carried in Zimbabwe aimed at analysing the frequency of
Nyanyadzi River floods using the Gumbel distribution. Extreme floods over recent years had washed
away fields, destroyed infrastructure and killed animals. It was hypothesised that Nyanyadzi flood flows
obey the Gumbel distribution. The scale and shape parameters of the distribution were estimated using
method of moments. Maximum instantaneous flow data covering 30 water years (1969-1999) for station
E119 on Nyanyadzi River were collected from the Zimbabwe National Water Authority (ZINWA). A Chi-
square
2
test revealed no significant (p =1.000) differences between recorded and predicted flood flows.
Due to goodness of fit of the Gumbel distribution, it was assumed to be appropriate for modelling
frequency of Nyanyadzi River floods. The magnitudes of the 100 and 200-year floods were estimated to be
276 and 310 m
3
/s respectively.

Key words. Flood, frequency; Gumbel distribution; Nyanyadzi; return period

Introduction

Nyanyadzi River catchment is one of regions in Zimbabwe where floods are causing enormous negative impacts
to both property and life. Over recent years, floods have killed animals, drowned cropland, destroyed
infrastructure and left people homeless. Outbreak of water-borne diseases, such as malaria and cholera is also
common during flood periods (Chenje et al., 1998; AGRITEX, 2000). Almost every year, a lot of resources are
invested for flood mitigation and protection using either structural (achieved by river training, storage dams and
weirs) and/or non-structural (achieved by means of flood forecasting and rescue operations) measures. However,
meteorological forecasts can only provide very short forecasts in an accurate form, which may not allow enough
time to reduce the impact of flood events. In addition, due to previous false alarms, people no longer take
forecasts seriously (Madamombe, 2005).

Given the above shortcomings of flood forecasting using rainfall data, this paper attempts to estimate return
periods associated with flood peaks of different magnitudes from recorded historical floods using statistical
method. The selected method is Gumbel extreme value distribution which is widely used for flood frequency
analysis.

Never Mujere / International Journal on Computer Science and Engineering (IJCSE)
ISSN : 0975-3397 Vol. 3 No. 7 July 2011 2774

Flood frequency analysis

Flood frequency analysis involves the fitting of a probability model to the sample of annual flood peaks
recorded over a period of observation, for a catchment of a given region. The model parameters established can
then be used to predict the extreme events of large recurrence interval (Pegram and Parak, 2004) Reliable flood
frequency estimates are vital for floodplain management; to protect the public, minimize flood related costs to
government and private enterprises, for designing and locating hydraulic structures and assessing hazards
related to the development of flood plains (Tumbare, 2000). Nevertheless, to determine flood flows at different
recurrence intervals for a site or group of sites is a common challenge in hydrology. Although studies have
employed several statistical distributions to quantify the likelihood and intensity of floods, none had gained
worldwide acceptance and is specific to any country (Law and Tasker, 2003).

Previous studies in Zimbabwe have estimated flood magnitude from catchment area, rainfall, shape slope and
length. It was observed that more than 90% (R
2
=0.93) of the variations in flood magnitude could be explained
by catchment area (Mitchell, 1974; Wurzel, 1986). However, rivers in the eastern highlands of the country were
not incorporated in the study because they showed unique behaviour. This study therefore applies the Gumbel
statistical distribution for flood frequency analysis. To the best of the authors knowledge, no previous studies in
the area have attempted to model flood discharges using the Gumbel distribution, a stochastic generating
structure that produce random outcomes. It was thus, hypothesized that Nyanyadzi River flood flows fit Gumbel
distribution model.

Materials and methods

Study area
Nyanyadzi River flows from its source in the humid eastern highlands (1 500 m), of Chimanimani district
westward into the dry Odzi River valley having an altitude of 400 m (Figure 1).


Fig. 1. Location of Nyanyadzi River system.

Nyanyadzi River and its tributaries drain a catchment area of 800 km
2
covering all the 4 agro-ecological regions
of the country. The river flows through large-scale commercial farmlands, resettlement and communal areas,
from a high rainfall area (1200 mm annually) into the dry, sparsely vegetated lowveld (400 mm annually). It
exhibits a simple flow regime (Mujere, 2006). Low flows occur during the dry months, from April to November
while high flows are experienced during the rain season, from December to March. River floods commonly
occur in January or February. They are dependent on the inter-tropical convergence low-pressure system and
tropical cyclones from the Indian Ocean (Tumbare, 2000; AGRITEX, 2000).

Never Mujere / International Journal on Computer Science and Engineering (IJCSE)
ISSN : 0975-3397 Vol. 3 No. 7 July 2011 2775


Data collection and analysis

Daily maximum instantaneous flow data (in m
3
/s) covering 30 water years of record for E119 gauging station on
Nyanyadzi River were obtained from the Zimbabwe National Water Authority (ZINWA). A water year in
Zimbabwe begins on October1
st
and ends on September 30
th
of the designated year. The flow recording station,
which was washed away by floods on 22 February 2000, was equipped with an automatic recorder.

Flow data were expressed in terms of exceedance probabilities and recurrence intervals. Denoting Q
i
as the
annual maximum flood in year i, the quantile Q
i
(F) is the value expected Q
i
to exceed with probability F, that
is, P(Q
i
Q
i
(F)) = F during the year of interest. Thus, there is a F% chance that Q Q(F). Conversely, there is
a (1F)% chance that X < Q(F). The return period of a flood, 1/(F) is the reciprocal of the probability of
exceedance in one year (Haan, 1977; Shaw, 1983).

Gumbel distribution

Gumbel distribution is a statistical method often used for predicting extreme hydrological events such as floods
(Zelenhasic, 1970; Haan, 1977; Shaw, 1983). In this study it has been applied for flood frequency analysis
because (a) peak flow data are homogeneous and independent hence lack long-term trends; (b) the river is less
regulated, hence is not significantly affected by reservoir operations, diversions or urbanization; and (c) flow
data cover a relatively long record (more than 10 years) and is of good quality (Mujere, 2006).

The equation for fitting the Gumbel distribution to observed series of flood flows at different return periods T is
(Sarma, 1999):

Q
t
= Q
av
+ K (1)
where
Q
t
denotes the magnitude of the T-year flood event, K is the frequency factor, Q
av
and are the mean and
the standard deviation of the maximum instantaneous flows respectively.

The frequency factor K is expressed as (Lehre, undated):

K=-6/(-ln(ln(T-ln(T-1))) (2)
where: =3.14, is the Euler constant (=0.5772) and ln is the natural logarithm.

The Chi-square
2
test was carried out to find the goodness of fit between the measured and predicted flood
flows. It was applied to test the hypothesis that the flood data fit the Gumbel distribution. After determining the
goodness of fit of the distribution to the flood discharges, flood magnitudes were computed for 0.5, 0.2, 0.1,
0.05, 0.04, 0.02, 0.01 and 0.005 exceedance probabilities.

Results

The maximum instantaneous flow of 233 m
3
/s was recorded at E119 in 1985 while the lowest flood flow of 8.83
m
3
/s was recorded in 1998. The 30-year mean instantaneous flood flow is 75.05 m
3
/s with a coefficient of
variability, CV of 85%. Measured and predicted flood flows show no significant (p=1.000) differences hence, a
goodness of fit of the Gumbel distribution (Figure 2.

Never Mujere / International Journal on Computer Science and Engineering (IJCSE)
ISSN : 0975-3397 Vol. 3 No. 7 July 2011 2776

0
50
100
150
200
250
1
9
6
9
1
9
7
1
1
9
7
3
1
9
7
5
1
9
7
7
1
9
7
9
1
9
8
2
1
9
8
4
1
9
8
6
1
9
8
8
1
9
9
0
1
9
9
2
1
9
9
4
1
9
9
6
1
9
9
8
Years
F
l
o
w

(
m
3
/
s
)
Measured Predicted

Fig. 2. Measured and predicted flood flows.

Due to the good relationship shown on Figure 3, the Gumbel model was used to predict flood flows for different
return periods (Table 1).

Table 1. Flood estimation using the Gumbel distribution.
Return period
T (years)
Flood flow
Q (m
3
/s)
2 64.54

5 121.07
10 158.49
20 194.39
25 205.78
50 240.86
100 275.68
200 310.37

Results have shown that Nyanyadzi River flood flows were variable during the 30 years of studied. The
2
test
revealed a satisfactorily fit between observed and estimated flood flow values. Hence, Gumbel distribution can
be used to predict frequency of floods.

Conclusion

Flood frequency analysis had been carried out for Nyanyadzi River using 30 years of peak flow data. The
outcome of the analysis clearly reveals the good capability of the Gumbel distribution function to predict river
flood magnitudes (Figure 2). There were no significant differences (p =1.000) between the predicted and
measured flow magnitudes. Hence, the model can be reliably applied to predict the occurrence of Nyanyadzi
River floods.

Never Mujere / International Journal on Computer Science and Engineering (IJCSE)
ISSN : 0975-3397 Vol. 3 No. 7 July 2011 2777


Acknowledgments
The author gratefully acknowledges the assistance and support of ZINWA staff for collecting, processing and
giving river flow data used in this study. I would like to recognize the valuable comments and suggestions made
by my friends.

References

[1] AGRITEX 2000. Assessment of the Impact of Cyclone Eline (February 2000) on the Food, Agriculture and Natural resource
Sector in Zimbabwe. AGRITEX, Harare.
[2] Haan, C.T. 1977. Statistical Methods in Hydrology. Iowa State University Press, Iowa.
[3] Law, G. S. and Tasker, G. D. (2003) Flood-Frequency Prediction Methods forUnregulated Streams of Tennessee, 2000. Water
Resources Investigations Report 03-4176, Nashville, Tennessee.
[4] Lehre, A. (undated) Methods of Streamflow Data Analysis.
http://www.humboldt.edu/~geodept/geology531/531_handouts/streamflow_data_
[5] Madamombe, E.K. 2005. Zimbabwe: Flood Management Practices: Selected Flood Prone Areas of Zambezi Basin, Zimbabwe
National Water Authority, Harare, Zimbabwe.
[6] Mitchell, T.B. 1974).A study of Rhodesian floods and proposed formulae, Rhod. Eng. 12(6), 199- 203.
[7] Mujere, N. 2006. Impact of river flow changes on irrigation agriculture: A case study of Nyanyadzi irrigation scheme in Chimanimani
district. MPhil Thesis, University of Zimbabwe, Harare.
[8] Pegram, G and Parak, M 2004. A review of the regional maximum flood and rational formula using geomorphological information and
observed floods, Water SA 30 (3) 377-388.
[9] Sarma, P. 1999. Flood risk zone mapping of Dikrong sub basin in Assam.
http://www.gisdevelopment.net/application/natural_hazards/floods
[10] Shaw, E.M. 1983. Hydrology in Practice. Van Nostrand Reinhold, UK
[11] Tumbare, M. J. 2000. Mitigating floods in Southern Africa. Paper presented at the 1
st
WARSFA/WaterNet Symposium: Sustainable
Use of Water Resources, 1-2 November, Maputo.
[12] Wurzel, P. 1986. Hydrology in Zimbabwe: The past and the future. Geographical Journal of Zimbabwe, 17, 1-18.
[13] Zelenhasic, E. 1970. Theoretical Probability Distributions for Flood Peaks. Colorado University Press, Colorado.
Never Mujere / International Journal on Computer Science and Engineering (IJCSE)
ISSN : 0975-3397 Vol. 3 No. 7 July 2011 2778